Hydrograph prediction in ungauged basins: Development of a closure relation for Hortonian runoff

Abstract

Hydrological forecasting and predictions under environmental change are often hampered by a lack of historical flow measurements and catchment physical data to characterize the system’s behaviour. This thesis presents a parsimonious semi-distributed rainfall-runoff modelling framework based on hydrological response units (HRUs) in which local-scale observable catchment characteristics can be directly used to parameterize the closure relations at the HRU scale. Thus, the modelling framework can potentially be used in ungauged basins where there is no sufficient data for ad-hoc parameter identification (i.e. calibration). The first part of thesis focuses on the development of Hortonian runoff closure relation for HRUs. An extensive data set of rainfall-runoff responses from hypothetical HRUs (6 x105 scenarios), generated from a detailed physically-based hydrological model, is used as a surrogate of real-world data sets to identify the form and parameters of the closure relation. These parameters are, in turn, related to local-scale HRU observables and HRU geometry for each scenario run, resulting in the parameter library to be used for the estimation of closure relation’s parameters for the HRUs outside the synthetic data sets. The closure relations show satisfactory performance in reproducing the observed hydrographs in the 16-km2 catchment in French Alps. Calibration of the closure relation against the observed discharge results in the saturated hydraulic conductivity that is comparable to the values obtained from plot measurements in the study catchment. Thus, the closure relation may be used without calibration if the local-scale observable HRU properties are correctly estimated. The second part of thesis investigates a technique for automated HRU delineation to support a model application at a large scale. This is done using a multiple-point geostatistics (MPS) technique in the context of geomorphological mapping. The MPS technique uses training images to derive the conditional relationships between occurrences of geomorphological types (HRUs) and a set of terrain attributes, consisting of four local morphometric properties and surrounding landforms at two locations. These relations are stored in a frequency database. In the mapping stage, a realization of a geomorphological class is assigned to the target mapping cell based on the probability function of landform class occurrences conditioned to the observed attributes, as retrieved from the frequency database. This technique is tested over 280 km2 in the Buëch catchment, France, using different sizes of training images. The best mapping accuracy (i.e. 51.2% of correct cell, evaluated against the field geomorphological map) can be obtained using training images covering 10% of total area. Small geomorphological features, i.e. hogbacks and glacises, are underrepresented in the automated maps due to sampling bias of these units in the training images. Using these automated geomorphological maps as the HRUs, thus, leads to substantial discharge underestimation, particularly in the dry period where hogbacks are the main runoff-contributing areas. However, this error has small effects for the predictions in the wet period because the catchment runoff is generated from many HRUs. The modelling framework presented in this thesis shows a promise and could serve be as a blueprint for predictions in ungauged basins.